Skip to content

Latest commit

Β 

History

History
81 lines (53 loc) Β· 3.09 KB

VISION.md

File metadata and controls

81 lines (53 loc) Β· 3.09 KB

Document Name: Project Vision Version: 0.1 Timestamp: 2026-02-09 Mon 16:33

_aSTT Project Vision

Purpose

Significantly increase efficiency with a modular desktop application (USB portable) for real-time speech recognition in 2–3 participant conversations, targeting physicians on Windows and macOS.

The app transcribes speech in real time, identifies who said what (diarization), and allows LLM processing of the text either immediately or after review/editing, plus export to other systems.


Target Audience

  • Physicians (e.g. in the practice or during consultations)
  • Users need: transcript with timestamps, speaker attribution, optional processing and editing (summaries, notes), and integration with existing platforms (EHR, other software)

Architecture: Modules

1. ASR Module (WhisperX)

2. LLM Module

  • Processes text from module 1 (transcript + diarization)
  • Local or various online LLM backends
  • Use cases: summaries, examination notes, suggested diagnoses, structured outputs for clinical documentation

3. Prompts, Dictionaries, Context Module (extensible)

  • Prompt management for the LLM
  • Dictionaries (domain terms, abbreviations)
  • Context (templates, presets by examination type / specialty)
  • Presets for LLM configuration
  • Input layer for quality and relevance of module 2 outputs

4. API Management & Export Module

  • API connection management to external platforms
  • Export of transcripts and LLM outputs to other systems (EHR, storage, other tools)

5. Settings & User Management Module

  • User accounts, profiles, preferences
  • App settings (audio, models, API keys, default prompts)
  • Security and privacy (local vs. cloud, data storage)
  • Licensing
  • Admin remote access and updates
  • WhisperX tuning options

Values

  • Privacy and Security
  • Modularity – modules can be developed and changed independently with clear interfaces
  • Physician usability – fast transcription, minimal friction, emphasis on diarization and structured output
  • LLM flexibility – choice of local vs. online model according to privacy and performance needs
  • Connectivity – export and API for integration into existing workflow and platforms
  • Development follows high-quality VibeCoding

Horizon

  1. Functional demo to validate Module 1 and 2 (core: ASR + diarization, basic online multi-LLM support). Optional time-limited trial.
  2. Long-term desktop product: Incremental development of all modules, including advanced prompts, dictionaries, and API. Sandbox.

This document is the basis for the project constitution (CONSTITUTION.md) and further specifications.

Previous Versions:

  • 0.0 β€” 2026-02-09 Mon 16:33 β€” Initial import (pre-versioned)